Across recent articles from Snowflake, KPMG and The Economist, the same theme emerges: the first big gains are likely to come from giving agents tightly controlled tasks such as gathering data, checking documents, monitoring signals, routing approvals and preparing recommendations.

The larger shift is not simply faster automation. It is a new model for financial work, one in which firms use stronger data foundations, clearer governance and human oversight to turn fragmented processes into more continuous workflows.

Snowflake argued that financial services is moving from using data mainly to understand problems to using data to take action. In a Wednesday (April 22) article on the Snowflake site, John Heisler wrote that banks, asset managers and insurers have spent years trying to collect, clean and connect data. That work helped teams make better decisions, but much of the process still depended on people pulling information from different systems.

Heisler said the next phase is agentic AI, where software agents can work across internal company data and outside market data in one governed environment. The goal is not to replace human judgment. It is to remove routine work so people can spend more time making higher-value decisions.

Heisler framed Snowflake’s “Ecosystem Agent Framework” as a way to make that shift practical in financial services. He cited the example of an investment analyst covering biopharma. Today, that analyst may need to check internal research, current holdings, portfolio exposure and market signals before forming a view. In Snowflake’s model, an agent could monitor those inputs continuously, flag relevant developments and prepare research notes with context for the analyst to review.

Heisler’s broader point is that useful agentic AI depends less on flashy demos and more on the data foundation beneath them. For financial firms, that means agents need access to data, strong controls, clear governance and the ability to act inside existing workflows.

The Strength Of Practical Use Cases

KPMG said agentic AI could help financial services firms turn automation from a limited tool into a broader operating model.

In an article attributed to KPMG, the firm argued that specialized AI agents can handle repeatable tasks, review emails and documents, analyze data sources and move work through process steps with less manual effort. KPMG framed the main value as productivity: faster workflows, lower process costs, fewer handoffs and better process quality.

The article also stressed that financial firms need controls before they scale these tools. KPMG said companies can use tailored agents to reflect their own business rules, data flows and approval steps, while giving business units more ability to build solutions without waiting on IT for every change. But that flexibility must sit inside clear governance, documentation, security and audit standards.

KPMG’s view is that agentic AI will be most useful when firms combine practical use cases with training, reusable components, approval mechanisms and a common platform for building and operating agents.

The Economist Checks In

Even The Economist said agentic AI is starting to reshape the manual work that sits behind many financial services processes. The article pointed to mortgage approvals as one example. A customer may deal with one loan officer, but behind the scenes, teams often handle dozens of steps across document review, compliance checks and underwriting.

The article said AI agents can take on many of these repeatable tasks, while people remain responsible for critical decisions and exceptions. That could help banks, insurers and other firms speed up audits, fraud detection, know your customer (KYC) reviews and credit assessments.

The article also made clear that the opportunity depends on careful execution. Financial firms already have large data sets and much of the cloud infrastructure needed to use AI agents, but they still need guardrails, investment and cultural change. Oracle modeled an agentic version of a mortgage workflow and projected that approval times could fall from 48 days to 38, while completed applications could rise 13%.

The article argued that firms should start with controlled use cases, protect customer data and measure results before scaling. Over time, the bigger prize may be redesigning work itself, not just making old processes faster.

For all PYMNTS AI and digital transformation coverage, subscribe to the daily AI and Digital Transformation Newsletters.